SemiAnalysis models 31.6% lower back-end network cost for a 1,024-H100 training cluster combining 2:1 leaf-to-spine oversubscription with a copper virtual-modular-switch layout, relative to NVIDIA's reference architecture (3 October 2024)
31.6% combined-design savingmodeledcontested
| Value kind | modeled — Interpret this value according to its displayed kind, scope, source, and as-of date. |
|---|---|
| Scope | The non-blocking virtual-modular-switch layout alone saves 24.9% against the same reference. Adding 2:1 oversubscription increases that saving by 6.7 percentage points, equivalent to about 8.9% of the non-blocking layout's cost (Guide derivation). Qualify an inference fabric separately from its traffic matrix and failed-state SLO. |
| As of | 2024-10-03 |
| Source | SemiAnalysis AI Neocloud Playbook |
| Review | checking…review by 2027-04-05 · fast cadence |
| Recorded changes | last 2026-10-05 · 3 revisions tracked |
| Claim id | inference-fabric-oversubscription-vs-1-1-non |
Where the guide uses it
← Full numbers register — every date-stamped figure in the guide, with revision history.